| Abstract |
The control of imaging parameters of a visual system is an important research area of computer vision. According to the theory of active vision, a visual system must constantly control its imaging parameters to aid in the performance of visual tasks. An important research topic in this context is gaze control, the control of mechanical parts and imaging parameters of the system. Gaze control can be subdivided into two mechanisms: gaze stabilization and gaze change. Gaze stabilization, which is the topic of this thesis, relates to the control of the degrees of freedom so that a world point can be stabilized on the image plane (which is either stable or moving with respect to the visual system). In particular, the gaze stabilization problem, is a problem of negative error feedback control. Thus, it requires the design of an error estimator, a controlled mechanical system and a control law which minimizes the error. In this thesis, a gaze stabilization system has been implemented for a stereo active vision head with four degrees of freedom. This system is based on the architecture proposed by D.J. Coombs \cite{Co92}. It consists of two cooperating subsystems, the vergence subsystem and the pursuit subsystem. The vergence subsystem estimates the error in the depth direction and, similarly, the pursuit subsystem estimates the positional error on the horopter (locus of points that stimulate exactly corresponding points). Both subsystems use these errors to direct the cameras, so that their optic axes instersect on the target's surface. For each subsystem we have implemented several techniques, already prosposed in the literature, in order to evaluate them in the context of gaze stabilization. For system control, we tested different ways by which errors can be distributed among the head's degrees of freedom in order to minimize the error of tracking. In order to evaluate the system, we developed a platform for the simulation of robot system kinematics. Using this platform, we simulated a stereoscopic head where a raytracer (Rayshade) plays the role of a camera. This platform can be used to represent more complex robot systems and can be easily extended to take into account their dynamics.
|
Υλοποίηση Συστήματος Σταθεροποίησης της Οπτικής Προσήλωσης Στερεοσκοπικής Κεφαλής Ενεργούς Ορασης
